The present invention relates primarily to the field of semiconductor and microstructure devices and their fabrication, and more particularly to a method of and apparatus for data storage and data transfer that limits or eliminates the need for motion of macroscopic electromechanical elements in one particular embodiment.
Much as the transistor replaced the vacuum tube, various other electronic functions are now being replaced by solid state devices. The attendant rate of miniaturization now allows one to make wires and other features that are as thin as 20 nm. Devices that have been produced range in size from micron to submicron. These microscopic sensors and actuators will be referred to hereinafter as microdevices. Important applications include microrobots for uses ranging from industrial production to microsurgery; optical devices for the generation, modulation and sensing of light; and complete microsystems for biological and chemical processing.
The focus has been on physical, chemical, and biological microstructures for use in microsensors, transducers, tactile and vibrational sensing arrays, and membranes. The capability required for fabricating such small devices has been enabled by advances in lithography, molecular-beam epitaxy, and metal/organic vapor deposition. These materials-processing and microfabrication techniques have been combined successfully with the technology of semiconductor manufacturing to produce various microdevices capable of functioning as electromagnetic sensors and actuators. Microdevices are disclosed in, for example, the proceedings of the "IEEE Solid-State Sensor and Actuator Workshop," Hilton Head Island, S.C., Jun. 6-9, 1988; "IEEE Workshop on Micro Electromechanical Systems," Salt Lake City, Utah, Feb. 20-22, 1989; and "IEEE Workshop on Micro Electro Mechanical Systems," Napa Valley, Calif., Feb. 11-14, 1990, all of which are incorporated herein by reference for all purposes.
The steps required for the fabrication of microdevices integrated with microelectronics on a chip or as a hybrid are well known to those skilled in the art. For example, the technique is being used to construct the entire transducer for a scanning tunneling microscope on a silicon chip, as described in Kenny et al., "A Micromachined Silicon Electron Tunneling Sensor," IEEE (1990), previously incorporated herein by reference for all purposes.
Electronic mass storage devices such as floppy disc drives, hard drives, and magnetic tape are well known. Three factors have provided the driving force in the quest for ever improved electronic mass storage: high information density, short access time, and long-term stability. The dominant technology for electronic-data mass storage over the past thirty years has been magnetic recording. The success of magnetic storage technology can be attributed at least in part to steady advances in providing the desired data capacity (for example, 10.sup.7 bits/cm.sup.2 on commercially available disks, and one order of magnitude larger on a recent demonstration disk) at a competitive price, albeit at the loss of speed, and to its substantially unlimited number of erasure cycles. While meeting with substantial success, difficulty has been encountered in providing a technology that performs satisfactorily with respect to all three factors. A typical design of the storage hierarchy involves tradeoffs, as a result of which most systems include a combination of (expensive) semiconductor memories, to provide a better match to the processor speed, and (slower) magnetic storage, to provide larger capacities for long-term storage.
The basic elements of a magnetic storage system comprise a magnetizable storage medium, a transducer that can write information to and/or read information from this medium, means for the medium and transducer to move with respect to each other, and suitable associated electronics. In a magnetic recording system, the transducer is called a head. The two most commonly used head technologies are based on inductive and flux-sensing methods and are described in, for example, Mee et al., Magnetic Recording Handbook, McGraw-Hill, 1990, which is incorporated herein by reference for all purposes.
The inductive head in a magnetic record/erase system includes a coil of wire wound around a magnetic core, and it relies on Faraday's law of induction. In the read process, the relative head/medium motion causes the head to pick up the time rate of change of the medium magnetization in the transition region, which induces a current in the coil. In the write process, a current passing through the coil creates a magnetic field in the head which is used to impress magnetized regions onto the storage medium. Unlike the inductive read transducer, the flux-sensitive read transducers do not require any motion relative to the storage medium. Flux-sensitive transducers include those based on a change in resistance (magnetoresistive effect), change in electric field (Hall effect), and modulating the reluctance of a ring core (flux gate). Much of the remaining description of read transducers herein relies on magnetoresistive transducers; this is not intended to be limiting. The magnetoresistive head relies on the changes in resistance of a magnetic material that accompanies a change in magnetization. It depends on the magnetic flux itself rather than on the rate of change of flux, as is the case with inductive heads; its output therefore depends only on the instantaneous fields from the media and is independent of the relative head/medium velocity or the time rate of change of the magnetic flux. The sensing element is biased with a magnetic field to optimize the linearity of its output. Many biasing schemes have been utilized, the largest class of which provides the biasing field through an auxiliary microstructure in close proximity to the magnetoresistive element. Furthermore, it is a read-only device, so that it has to be combined with an inductive write head. Magnetoresistive tape heads are available commercially, disk heads, on a demonstration basis.
Despite its dominant position, magnetic-storage technology suffers from several basic problems, arising mainly not from storage itself but rather from the present method of transfer of data between mass storage and the computer by means of moving heads. Disadvantages include:
(1) The relatively slow speed of access. It takes on the order of tens of milliseconds to transfer a block from disk storage. PA1 (2) It is vulnerable to shock and vibration. PA1 (3) Materials problems arising from the relative head/medium motion. Materials choices generally represent a compromise between the desired electrical performance and tribological (wear and friction) constraints. PA1 (4) The practical limitation on the density of magnetic storage is currently set by the size of the read/write head; it is presently two orders of magnitude short of the theoretical limitation in many embodiments. In principle, each magnetic domain can encode one bit. In practice, locating or addressing individual domains presents problems that have proven insurmountable to date, because of the size of the head. Practical considerations normally dictate that in magnetic-recording systems based on prior-art head/disk technology each bit contains a large number of domains, which precludes reaching the theoretical storage capacity. PA1 (5) The magnetization in the direction normal to the surface of the medium falls off exponentially with distance. There is a corresponding loss of sensing signal with increased head/medium spacing. Designs thus represent a compromise between the close spacing essential for high-density storage and the need to maintain the stability required to avoid contacts. PA1 (1) Seek time (or access motion time) is the time required for the access arm to position its read/write head over the proper track. PA1 (2) Rotational delay (or latency)is the time it takes for the rotating disk to bring the desired sector under the access arm. PA1 (3) Data transfer time (or data movement time) is the time required to transfer data between the disk and main memory. PA1 access time=seek time+rotational delay+data-transfer time PA1 positioning time=seek time+rotational delay PA1 (1) Information (audio, video, or data) to be recorded magnetically is encoded as a time-varying electrical signal. PA1 (2) The signal current with the encoded pattern is applied to the writing-head windings. PA1 (3) This current magnetizes the head. PA1 (4) The fringe magnetic field from the head creates, on the moving medium, a spatially varying pattern of magnetization that reproduces the pattern encoded in the electrical signal. PA1 (a) In the read process, the motion of the windings in the head through the magnetic-field lines from the storage medium generates current in these windings. This is the fundamental mechanism underlying the actual transfer of a given bit between the medium and an inductive head. This aspect of the motion does not come into play for a magnetoresistive head. PA1 (b) The relative motion of the head along the track accesses the data in the sector serially. In the recording process, this produces a magnetization pattern according to the input current applied to the head. If the input signal is at a frequency f and the medium is moving at a relative velocity v, a magnetization pattern (0s and 1s) will be recorded at a fundamental wavelength given by EQU .lambda.=v/f. (1)
Computer applications require large amounts of data transfer between internal computer memory and an external storage device, such as a disk. There is generally a large disparity between the internal processing speed of the computer and speed of input/output (I/O). Typical computer instruction times range from the order of a microsecond down to tens of nanoseconds; a typical operation to transfer a sector of data is of the order of tens of milliseconds. Although comparison between the transfer of a sector and a single instruction is clearly not a direct measure of the relative times required for I/O and for processing, it is well known from practical experience that typical I/O times in data-transfer-intensive applications can be several orders of magnitude longer than typical CPU processing times. This large speed disparity between processing and I/O reflects the vast difference between the time constants that characterize these two functions. CPU processing time is dictated by transistor switching times, which are orders of magnitude shorter than the characteristic times of motion of the macroscopic electromechanical parts essential to prior-art head/disk mass storage systems. In such applications, I/O, not processing time, is the limiting factor in throughput. Overcoming this critical limitation in the overall speed of computing is one of the major problems in computer-system design.
Magnetic disks store data in concentric circles called tracks. Digital data are stored serially around the track. Each track is divided into sectors. A sector is a group of contiguous bits, which are generally transferred between memory and disk in one I/O operation. The data are accessed by read/write heads mounted on the ends of access arms. In most disk units, the heads are positioned over a given sector by a combination of two mechanical motions: the disk rotates to provide angular position, and the access arm moves radially to provide radial position. The combination of disk rotation and access-arm motion allows the head to be positioned over any point (any sector of a disk and track) of a disk.
Accordingly, the time required to move data between a disk and internal computer memory (access time) has three major components. These components represent three separate actions in the data storage and retrieval process.
The access time required to read (or write) on the disk is the sum of the three times:
The sum of the average seek time and rotational delay is referred to hereinafter as positioning time:
As indicative of typical times, the HP 7935H has an average seek time of 24 ms, rotational delay of 11.1 ms, and data-transfer time of 1.0 ms for a kilobyte sector; the corresponding times for the IBM 3380 are 16 ms, 8.3 ms, and 0.33 ms, respectively. The average positioning time, which is seen to be 30-80 times longer than the actual data-transfer time from a sector, is clearly the limiting factor in access time.
The technology of data storage and retrieval by prior-art magnetic-storage systems relies on the relative motion of the head and the storage medium. During transport of the recording medium past the head the relative motion of the two permits writing or reading; in general, this motion causes a transfer between a temporal signal in the read/write head and a recorded spatial pattern in the medium.
The fundamental process, in which temporal input data are translated into a recorded spatial magnetization pattern in the medium during a write, involves several steps.
The reading process uses either the same head or another head to reconvert the recorded magnetization pattern into a time-varying electrical signal that can be amplified to a useful level, for example, to feed data to a computer or to drive a loudspeaker or a receiver.
The rotation of the disk past the head actually serves a dual role in these devices.
In general, all temporal signal variations are translated into spatial variations by the relation EQU x=vt, (2)
where x denotes the pattern coordinate along the medium and t is the temporal coordinate of the input signal. This aspect of the head/medium motion is required for both inductive and magnetoresistive heads.
In addition to their dependence on the speed v, the phenomena associated with the write process depend, among other parameters, on the head-medium separation d (otherwise known as the flying height). See, for example, "Special Section on Magnetic Information Storage Technology," Proceedings of the IEEE, November 1986, Vol. 74, which is incorporated herein by reference for all purposes. Apart from differences attendant to the head functioning as a sensor rather than as an actuator, the signal produced when the medium is read is a function of most of these same parameters, including v and d.
Semiconductor random access memories (RAMs) are also well know to those of skill in the art. A RAM generally comprises a set of memory cells integrated on a chip with a number of peripheral circuits. RAMs are described in, for example, Porat et al., Introduction to Digital Techniques, John Wiley, 1979, which is incorporated herein by reference for all purposes. In general, RAM circuits perform several functions, including addressing (selection of specific locations for access), providing power, fanout (transmission of a signal to a multiplicity of directions), and conditioning required to generate a useable output signal. In RAM memories, the addressing scheme permits direct access to the desired cell, with access time independent of the cell location. Selected portions are then extracted for use. RAMs are generally fast enough to be compatible with a CPU, but they are generally too expensive to be used for mass storage. Further, RAMs are generally volatile in the sense that a source of power must be provided to refresh the memory periodically. They cannot, therefore, be used for long-term storage. One alternative to RAM includes ROM such as EPROMs (Electronically Programmable Read Only Memory). While such memories do not require a refresh cycle, they have the obvious disadvantage of being progammable only once. Other nonvolatile semiconductor memories that can be written repeatedly, such as EAROM (Electrically Alterable Read Only Memory) or EEROM (Electrically Erasable Read Only Memory), do not provide nearly the reliability of magnetic memories for long-term storage.
The scanning tunneling microscope (STM) is an instrument used for measuring surface properties. It comprises a sharp needle, usually made of tungsten, that can probe the electronic structure of conducting surfaces by means of the tunneling effect. The probe is placed in close proximity to the surface and physically scanned over it. The instrument provides a tool for characterizing static surface properties of, for example, conductors. It is unique in providing surface information on an atomic scale and has opened opportunities in applications ranging from biological systems to telerobotics. Derivatives of the STM have extended the capabilities of tunneling sensors to the measurement of nonconducting as well as conducting surfaces.
There are however whole classes of phenomena involving transient surface effects that are of wide interest but which take place on vastly shorter time scales than the time needed to scan a surface physically by an STM probe. Many of these effects bear directly on the development of new electronic materials, devices and circuits. There is thus a need to develop novel methods for the measurement of dynamic surface effects that derives from the confluence of several factors: the importance of their application; the facts that surface properties are in general significantly different from and not nearly as well known as bulk properties; and the difficulty of measuring dynamic surface effects on an atomic scale by conventional methods.
From the above it is seen that improved transducer arrays are desired.